My first attempt at AI video for TikTok was a beautiful 16:9 shot that I cropped to vertical afterward. The composition was perfect in the editor and useless on a phone — the subject's head was clipped, the interesting half of the frame was gone, and the caption overlay covered whatever was left. I'd built a landscape video and then punished it for not being portrait.
The fix wasn't a better prompt. It was generating vertical from the start, and treating the first second as the entire job. This guide walks through the workflow I use now: the settings, the timing, the platform specs, and the mistakes that kill a vertical AI video before anyone sees frame two.
The short answer
To make AI video for TikTok and Shorts that actually performs, set the aspect ratio to 9:16 before you generate — not after — target 1080x1920, keep each clip in the 5 to 8 second range, and build a visual hook that lands inside the first second. Then stitch two or three clips into a 15 to 30 second cut, which is where engagement peaks on both platforms.
Everything below is the detail behind that sentence.
Why vertical has to be a generation setting, not an edit
Aspect ratio is a compositional anchor, not a container. When you tell the model 9:16 up front, it frames the subject for a tall canvas: head and shoulders sit in the upper third, the action stays center-column, and there's breathing room where the UI will sit. Ask for 16:9 and crop later and you're throwing away roughly 60% of the horizontal frame — usually the wrong 60%.
Kling 3.0 supports 16:9, 9:16, and 1:1 natively, so there's no reason to generate the wrong shape. If you want the deeper reasoning on ratio choice, we covered it in the Kling 3.0 aspect ratio guide.
Rule of thumb: if the video is going to a phone-first feed, pick 9:16 in the generator, never in the editor.
There's a second reason. Both TikTok and YouTube overlay UI on your video — captions, username, sound label, the right-hand action rail, and on Shorts the progress bar. Roughly the bottom 15% and the right 10% of the frame are contested space. A vertical ai video generated at 9:16 gives you enough headroom to keep your subject out of that zone. A cropped landscape shot almost never does.
Platform specs you should actually design against
| Spec | TikTok | YouTube Shorts |
|---|---|---|
| Aspect ratio | 9:16 | 9:16 (square also accepted) |
| Resolution | 1080x1920 | 1080x1920 |
| Max length | 3 minutes for in-app recording; longer for uploads | 3 minutes to still count as a Short |
| Best-performing length | 15–30 seconds | 15–30 seconds |
| Format | MP4 / MOV | MP4, H.264 + AAC |
| Safe zone | Keep subject clear of bottom 15%, right 10% | Same, plus progress bar at the very bottom |
The length row is the one people misread. "Up to 3 minutes" is a ceiling, not a target. Since December 2025, YouTube classifies vertical uploads up to three minutes as Shorts — but retention data still favours the 15–30 second band on both platforms. Longer is permitted; longer is rarely better.
The workflow, step by step
Here's the sequence I run for every ai short form video, from blank prompt to upload-ready file.
- Write the hook as a shot, not a sentence. Decide what's visibly happening at 0:00. A face already mid-reaction, a hand already moving, a subject already walking toward camera. Static establishing shots are death in a feed.
- Set 9:16 and your duration before prompting. In Kling 3 AI, the ratio and length are set alongside the prompt. Kling 3.0 supports 3 to 15 seconds per generation; I default to 5 for a hook clip and 8 for anything with a story beat.
- Prompt for a center-column subject. Add framing language explicitly — "medium shot, subject centered, head in upper third, vertical composition." The model respects framing instructions far more reliably than it infers them.
- Pick your resolution mode. Kling 3.0 offers 720p and 1080p, with native-audio and no-native-audio variants. For social, generate 1080p; the feed compresses hard and 720p source shows it.
- Decide on native audio early. Kling 3.0 can generate native audio with lip-sync. If your video has a speaking character, use it. If you're scoring to a trending sound, generate silent and save the credits.
- Generate two or three clips, then cut. A single 8-second clip is a fragment. Two 8-second clips with a hard cut between them is a video. The cut itself is a retention device.
- Export at 1080x1920 MP4 and add captions in-platform. Native captions get better distribution than burned-in text, and they let you keep the generated frame clean.
Clip length: what to pick and why
| Clip use | Length | Why |
|---|---|---|
| Pure hook / loop | 3–5s | Loops cleanly, cheap to iterate, high replay count |
| Single-beat demo | 5–8s | Enough for one action to complete without dead air |
| Story or dialogue beat | 8–15s | Room for setup and payoff; use native audio here |
| Full upload | 15–30s total | Stitched from 2–4 clips; the engagement sweet spot |
Rule of thumb: generate short clips and assemble, don't generate one long clip and trim. Three 6-second generations give you three chances at a usable take and a natural cut point between each. One 15-second generation gives you one take, and if second nine is mushy you've wasted the whole thing.
The four mistakes that kill vertical AI video
Cropping landscape to portrait. Covered above, but it's the number one cause of "why does this look cheap." Regenerate at 9:16 instead.
Burying the hook. A three-second slow push-in toward a subject is cinematic and completely wrong for a feed. Start on the interesting frame. If your best moment is at 0:04, cut the first four seconds.
Ignoring the safe zone. Generate a gorgeous centered composition, then watch the TikTok action rail land on your subject's face. Prompt for the subject slightly left of center, or leave headroom.
Treating each clip as a standalone. Character consistency across clips is what separates a video from a slideshow. Kling 3.0 improved character consistency specifically for multi-shot work — reuse the same reference image across generations rather than re-rolling from text each time.
If you're making anything promotional rather than editorial, the structural rules differ enough that it's worth reading the AI video ads guide alongside this one.
Frequently asked questions
How do I make AI video for TikTok? Set the aspect ratio to 9:16 in your generator before prompting, target 1080x1920, generate clips of 5 to 8 seconds each, and stitch two or three together into a 15 to 30 second upload. Prompt explicitly for vertical framing with the subject in the center column.
What aspect ratio should ai youtube shorts use? 9:16 at 1080x1920. YouTube also accepts square, but 9:16 fills the screen and is what the format is designed around. Vertical uploads up to three minutes are classified as Shorts.
Can Kling 3.0 generate vertical ai video natively? Yes. Kling 3.0 supports 16:9, 9:16, and 1:1 as generation-time settings, at 720p or 1080p, with clip durations from 3 to 15 seconds. You can try it in the browser at Kling 3 AI.
How long should an ai short form video be? The platforms allow up to three minutes, but engagement data consistently favours 15 to 30 seconds. Build that from multiple short generated clips rather than one long one.
Do I need audio in the generated clip? Only if a character speaks. Kling 3.0 offers native audio with lip-sync, which is worth it for dialogue. If you're pairing the video with a trending sound, generate without native audio and score it in-platform.
Will AI-generated video get labelled or restricted? Both TikTok and YouTube require disclosure of realistic AI-generated content and apply their own labels. Policies here change frequently — check each platform's current synthetic media rules before you post at scale rather than relying on any guide, including this one.
The bottom line
Vertical AI video isn't landscape video wearing a different crop. It's a different composition, a different clip length, and a different relationship with the first second. Set 9:16 at generation time, target 1080x1920, keep clips between 5 and 8 seconds, assemble 15 to 30 second cuts, and keep your subject out of the UI zone. That's most of the job.
The fastest way to internalize it is to generate one 9:16 clip and one 16:9 clip from the same prompt and look at both on your phone. Open Kling 3 AI, run the pair, and the difference will be obvious in about four seconds — which, conveniently, is roughly how long you have on the feed anyway.
Sources
- Kling VIDEO 3.0 Model Guide — Kling AI official: official model specs including resolution modes (720p/1080p), native-audio variants, and generation settings.
- Kling 3.0 15s Video: Narrative Control & Custom Duration — Kling AI official: official confirmation of the 3-to-15-second flexible duration range.
- Understand three-minute YouTube Shorts — YouTube Help: official rule that square or vertical uploads up to three minutes are classified as Shorts, effective December 8, 2025.
- Social media video specs guide — Sprout Social: maintained reference for TikTok and Shorts dimensions, formats, and length limits.
A note on sourcing: platform specs, length limits, and AI-content disclosure policies change often, and Kling model capabilities change with each release. The figures here reflect mid-2026 information from the sources above. Verify current specs on the official TikTok, YouTube, and Kling AI pages before building a production pipeline, and check each platform's own synthetic media policy rather than treating this guide as authoritative on labelling requirements.
